The performance of 3 candidate sets of generic Head-Related Transfer Functions (HRTFs) was evaluated in terms of differences in their ability to create distinct elevation angles for virtual sources located on a listener’s median plane. In comparison to directional errors observed when 6 listeners were presented with sources processed using their own individually measured Head-Related Impulse Responses (HRIRs), the errors associated with the 3 alternative candidate sets were tested for seven virtual source target elevation angles on the median plane, spanning the upper hemifield with 30 deg. resolution from front to back. One candidate set was that of a selected subject’s HRIRs that had been modified through a customization procedure designed to provide optimal elevation performance for that subject (which was the selected ‘representative’ subject denoted ‘CH’ in Hwang, et al. [12]). Another candidate set was that formed by taking the mean over the optimized HRIRs of 9 subjects that resulted from that same customization procedure. The third set was generated by taking the mean over the HRIRs of 43 subjects found in the CIPIC database. Performance under identical conditions was also observed for sources convolved with the appropriate individually customized HRIR for each of the 6 listener’s, and also using a familiar KEMAR dataset to serve as a reference. Localization performance was quantified using the 3 error metrics of absolute polar angle error, vertical polar angle error, and front-back confusion rate. Only in terms of front-back confusion rate was performance shown to be inferior for virtual sources processed using the mean of the HRIRs from the CIPIC dataset when compared to either those processed using the representative subject’s customized HRIRs, or using mean of customized HRIRs of 9 subjects. Although the study investigated virtual sources varying in elevation on the median-plane only, the results may generalize to the whole upper hemifield.
https://www.aes.org/e-lib/browse.cfm?elib=15551
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.
Learn more about the AES E-Library
Start a discussion about this paper!
